How Swiss Railways Uses AI to Detect Track Defects

SBB's diagnostic trains inspect around 2,000 kilometres of track every month. The rolling inspection programme gives Swiss Federal Railways a faster way to identify damage across the network, while reducing the need for staff to examine long stretches of railway on foot.

Two specially equipped trains travel at speeds of up to 120 kilometres per hour, recording the condition of the track as they move. Cameras capture detailed images, while measurement systems monitor the geometry of the rails. Artificial intelligence then reviews the visual data and flags signs of damage for further assessment.

The scale matters in a country where trains run through dense urban areas, Alpine valleys and exposed mountain routes. Track conditions can change because of heavy use, weather, construction activity and ordinary wear. Regular automated inspections help SBB build a detailed picture of those changes across the network.

According to Swiss public broadcaster SRF, the trains detect as many as 1,500 defects per month, most of them minor. That figure represents a stream of maintenance information rather than a list of immediate emergencies. Each finding must be assessed, prioritised and, where necessary, repaired by railway specialists.

The inspection system combines high speed cameras with precise measurements of rail geometry. Those tools examine different aspects of the infrastructure and give SBB more than a simple photographic record.

Track geometry equipment can detect minute deviations in the rails. Even small changes in alignment, level or spacing can provide an early indication that a section needs attention. Cameras add a visual layer, capturing images that AI systems can screen for visible faults and irregularities.

The software has been trained to recognise particular defects in defined parts of the railway. It compares new images and measurements with the patterns it has learned, then highlights possible problems for human review. This process allows specialists to concentrate on flagged locations instead of searching every metre manually.

SBB started testing AI for track inspection about a decade ago. Early results were disappointing because the system generated too many errors. Improvements in AI have since increased its accuracy and reliability, according to SRF. The technology still works as part of an inspection chain. Data collection, automated analysis, expert verification and physical maintenance all remain connected.

AI flags defects, but it does not inspect every part of the railway system. Its effectiveness depends on the areas covered by its training and the equipment installed on the diagnostic trains.

The system is designed to recognise specific problems in specific parts of the track. That focus improves performance, but it also creates boundaries. Damage outside those areas may go unnoticed. SRF's reporting gives overhead line masts as one example of infrastructure that the track inspection system will not necessarily detect.

That limitation makes human judgement essential. Railway staff must review the findings, investigate conditions that fall outside the software's field of vision and decide how urgently SBB needs to respond. Workers also carry out the repairs, from correcting track geometry to replacing damaged components.

The result is a division of labour rather than a fully automated railway. Machines collect large volumes of data at speed. Software sorts and prioritises the information. Experienced railway workers interpret the warnings and intervene on the ground. For passengers, the practical benefit lies in earlier and more consistent identification of defects, provided that the follow up remains thorough.